slab.c 111 KB
Newer Older
1
// SPDX-License-Identifier: GPL-2.0
Linus Torvalds's avatar
Linus Torvalds committed
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
/*
 * linux/mm/slab.c
 * Written by Mark Hemment, 1996/97.
 * (markhe@nextd.demon.co.uk)
 *
 * kmem_cache_destroy() + some cleanup - 1999 Andrea Arcangeli
 *
 * Major cleanup, different bufctl logic, per-cpu arrays
 *	(c) 2000 Manfred Spraul
 *
 * Cleanup, make the head arrays unconditional, preparation for NUMA
 * 	(c) 2002 Manfred Spraul
 *
 * An implementation of the Slab Allocator as described in outline in;
 *	UNIX Internals: The New Frontiers by Uresh Vahalia
 *	Pub: Prentice Hall	ISBN 0-13-101908-2
 * or with a little more detail in;
 *	The Slab Allocator: An Object-Caching Kernel Memory Allocator
 *	Jeff Bonwick (Sun Microsystems).
 *	Presented at: USENIX Summer 1994 Technical Conference
 *
 * The memory is organized in caches, one cache for each object type.
 * (e.g. inode_cache, dentry_cache, buffer_head, vm_area_struct)
 * Each cache consists out of many slabs (they are small (usually one
 * page long) and always contiguous), and each slab contains multiple
 * initialized objects.
 *
 * This means, that your constructor is used only for newly allocated
Simon Arlott's avatar
Simon Arlott committed
30
 * slabs and you must pass objects with the same initializations to
Linus Torvalds's avatar
Linus Torvalds committed
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
 * kmem_cache_free.
 *
 * Each cache can only support one memory type (GFP_DMA, GFP_HIGHMEM,
 * normal). If you need a special memory type, then must create a new
 * cache for that memory type.
 *
 * In order to reduce fragmentation, the slabs are sorted in 3 groups:
 *   full slabs with 0 free objects
 *   partial slabs
 *   empty slabs with no allocated objects
 *
 * If partial slabs exist, then new allocations come from these slabs,
 * otherwise from empty slabs or new slabs are allocated.
 *
 * kmem_cache_destroy() CAN CRASH if you try to allocate from the cache
 * during kmem_cache_destroy(). The caller must prevent concurrent allocs.
 *
 * Each cache has a short per-cpu head array, most allocs
 * and frees go into that array, and if that array overflows, then 1/2
 * of the entries in the array are given back into the global cache.
 * The head array is strictly LIFO and should improve the cache hit rates.
 * On SMP, it additionally reduces the spinlock operations.
 *
Andrew Morton's avatar
Andrew Morton committed
54
 * The c_cpuarray may not be read with enabled local interrupts -
Linus Torvalds's avatar
Linus Torvalds committed
55
56
57
58
 * it's changed with a smp_call_function().
 *
 * SMP synchronization:
 *  constructors and destructors are called without any locking.
59
 *  Several members in struct kmem_cache and struct slab never change, they
Linus Torvalds's avatar
Linus Torvalds committed
60
61
62
63
64
65
66
67
68
69
70
71
 *	are accessed without any locking.
 *  The per-cpu arrays are never accessed from the wrong cpu, no locking,
 *  	and local interrupts are disabled so slab code is preempt-safe.
 *  The non-constant members are protected with a per-cache irq spinlock.
 *
 * Many thanks to Mark Hemment, who wrote another per-cpu slab patch
 * in 2000 - many ideas in the current implementation are derived from
 * his patch.
 *
 * Further notes from the original documentation:
 *
 * 11 April '97.  Started multi-threading - markhe
72
 *	The global cache-chain is protected by the mutex 'slab_mutex'.
Linus Torvalds's avatar
Linus Torvalds committed
73
74
75
76
77
78
 *	The sem is only needed when accessing/extending the cache-chain, which
 *	can never happen inside an interrupt (kmem_cache_create(),
 *	kmem_cache_shrink() and kmem_cache_reap()).
 *
 *	At present, each engine can be growing a cache.  This should be blocked.
 *
79
80
81
82
83
84
85
86
87
 * 15 March 2005. NUMA slab allocator.
 *	Shai Fultheim <shai@scalex86.org>.
 *	Shobhit Dayal <shobhit@calsoftinc.com>
 *	Alok N Kataria <alokk@calsoftinc.com>
 *	Christoph Lameter <christoph@lameter.com>
 *
 *	Modified the slab allocator to be node aware on NUMA systems.
 *	Each node has its own list of partial, free and full slabs.
 *	All object allocations for a node occur from node specific slab lists.
Linus Torvalds's avatar
Linus Torvalds committed
88
89
90
91
 */

#include	<linux/slab.h>
#include	<linux/mm.h>
92
#include	<linux/poison.h>
Linus Torvalds's avatar
Linus Torvalds committed
93
94
95
96
97
#include	<linux/swap.h>
#include	<linux/cache.h>
#include	<linux/interrupt.h>
#include	<linux/init.h>
#include	<linux/compiler.h>
98
#include	<linux/cpuset.h>
99
#include	<linux/proc_fs.h>
Linus Torvalds's avatar
Linus Torvalds committed
100
101
102
103
104
105
106
#include	<linux/seq_file.h>
#include	<linux/notifier.h>
#include	<linux/kallsyms.h>
#include	<linux/cpu.h>
#include	<linux/sysctl.h>
#include	<linux/module.h>
#include	<linux/rcupdate.h>
107
#include	<linux/string.h>
108
#include	<linux/uaccess.h>
109
#include	<linux/nodemask.h>
110
#include	<linux/kmemleak.h>
111
#include	<linux/mempolicy.h>
Ingo Molnar's avatar
Ingo Molnar committed
112
#include	<linux/mutex.h>
113
#include	<linux/fault-inject.h>
Ingo Molnar's avatar
Ingo Molnar committed
114
#include	<linux/rtmutex.h>
115
#include	<linux/reciprocal_div.h>
116
#include	<linux/debugobjects.h>
117
#include	<linux/memory.h>
118
#include	<linux/prefetch.h>
119
#include	<linux/sched/task_stack.h>
Linus Torvalds's avatar
Linus Torvalds committed
120

121
122
#include	<net/sock.h>

Linus Torvalds's avatar
Linus Torvalds committed
123
124
125
126
#include	<asm/cacheflush.h>
#include	<asm/tlbflush.h>
#include	<asm/page.h>

127
128
#include <trace/events/kmem.h>

129
130
#include	"internal.h"

131
132
#include	"slab.h"

Linus Torvalds's avatar
Linus Torvalds committed
133
/*
134
 * DEBUG	- 1 for kmem_cache_create() to honour; SLAB_RED_ZONE & SLAB_POISON.
Linus Torvalds's avatar
Linus Torvalds committed
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
 *		  0 for faster, smaller code (especially in the critical paths).
 *
 * STATS	- 1 to collect stats for /proc/slabinfo.
 *		  0 for faster, smaller code (especially in the critical paths).
 *
 * FORCED_DEBUG	- 1 enables SLAB_RED_ZONE and SLAB_POISON (if possible)
 */

#ifdef CONFIG_DEBUG_SLAB
#define	DEBUG		1
#define	STATS		1
#define	FORCED_DEBUG	1
#else
#define	DEBUG		0
#define	STATS		0
#define	FORCED_DEBUG	0
#endif

/* Shouldn't this be in a header file somewhere? */
#define	BYTES_PER_WORD		sizeof(void *)
David Woodhouse's avatar
David Woodhouse committed
155
#define	REDZONE_ALIGN		max(BYTES_PER_WORD, __alignof__(unsigned long long))
Linus Torvalds's avatar
Linus Torvalds committed
156
157
158
159
160

#ifndef ARCH_KMALLOC_FLAGS
#define ARCH_KMALLOC_FLAGS SLAB_HWCACHE_ALIGN
#endif

161
162
163
164
165
166
167
168
169
#define FREELIST_BYTE_INDEX (((PAGE_SIZE >> BITS_PER_BYTE) \
				<= SLAB_OBJ_MIN_SIZE) ? 1 : 0)

#if FREELIST_BYTE_INDEX
typedef unsigned char freelist_idx_t;
#else
typedef unsigned short freelist_idx_t;
#endif

170
#define SLAB_OBJ_MAX_NUM ((1 << sizeof(freelist_idx_t) * BITS_PER_BYTE) - 1)
171

Linus Torvalds's avatar
Linus Torvalds committed
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
/*
 * struct array_cache
 *
 * Purpose:
 * - LIFO ordering, to hand out cache-warm objects from _alloc
 * - reduce the number of linked list operations
 * - reduce spinlock operations
 *
 * The limit is stored in the per-cpu structure to reduce the data cache
 * footprint.
 *
 */
struct array_cache {
	unsigned int avail;
	unsigned int limit;
	unsigned int batchcount;
	unsigned int touched;
189
	void *entry[];	/*
Andrew Morton's avatar
Andrew Morton committed
190
191
192
193
			 * Must have this definition in here for the proper
			 * alignment of array_cache. Also simplifies accessing
			 * the entries.
			 */
Linus Torvalds's avatar
Linus Torvalds committed
194
195
};

Joonsoo Kim's avatar
Joonsoo Kim committed
196
197
198
199
200
struct alien_cache {
	spinlock_t lock;
	struct array_cache ac;
};

201
202
203
/*
 * Need this for bootstrapping a per node allocator.
 */
204
#define NUM_INIT_LISTS (2 * MAX_NUMNODES)
205
static struct kmem_cache_node __initdata init_kmem_cache_node[NUM_INIT_LISTS];
206
#define	CACHE_CACHE 0
207
#define	SIZE_NODE (MAX_NUMNODES)
208

209
static int drain_freelist(struct kmem_cache *cache,
210
			struct kmem_cache_node *n, int tofree);
211
static void free_block(struct kmem_cache *cachep, void **objpp, int len,
212
213
			int node, struct list_head *list);
static void slabs_destroy(struct kmem_cache *cachep, struct list_head *list);
214
static int enable_cpucache(struct kmem_cache *cachep, gfp_t gfp);
215
static void cache_reap(struct work_struct *unused);
216

217
218
219
220
221
static inline void fixup_objfreelist_debug(struct kmem_cache *cachep,
						void **list);
static inline void fixup_slab_list(struct kmem_cache *cachep,
				struct kmem_cache_node *n, struct page *page,
				void **list);
222
223
static int slab_early_init = 1;

224
#define INDEX_NODE kmalloc_index(sizeof(struct kmem_cache_node))
Linus Torvalds's avatar
Linus Torvalds committed
225

226
static void kmem_cache_node_init(struct kmem_cache_node *parent)
227
228
229
230
{
	INIT_LIST_HEAD(&parent->slabs_full);
	INIT_LIST_HEAD(&parent->slabs_partial);
	INIT_LIST_HEAD(&parent->slabs_free);
231
	parent->total_slabs = 0;
232
	parent->free_slabs = 0;
233
234
	parent->shared = NULL;
	parent->alien = NULL;
235
	parent->colour_next = 0;
236
237
238
239
240
	spin_lock_init(&parent->list_lock);
	parent->free_objects = 0;
	parent->free_touched = 0;
}

Andrew Morton's avatar
Andrew Morton committed
241
242
243
#define MAKE_LIST(cachep, listp, slab, nodeid)				\
	do {								\
		INIT_LIST_HEAD(listp);					\
244
		list_splice(&get_node(cachep, nodeid)->slab, listp);	\
245
246
	} while (0)

Andrew Morton's avatar
Andrew Morton committed
247
248
#define	MAKE_ALL_LISTS(cachep, ptr, nodeid)				\
	do {								\
249
250
251
252
	MAKE_LIST((cachep), (&(ptr)->slabs_full), slabs_full, nodeid);	\
	MAKE_LIST((cachep), (&(ptr)->slabs_partial), slabs_partial, nodeid); \
	MAKE_LIST((cachep), (&(ptr)->slabs_free), slabs_free, nodeid);	\
	} while (0)
Linus Torvalds's avatar
Linus Torvalds committed
253

254
255
#define CFLGS_OBJFREELIST_SLAB	((slab_flags_t __force)0x40000000U)
#define CFLGS_OFF_SLAB		((slab_flags_t __force)0x80000000U)
256
#define	OBJFREELIST_SLAB(x)	((x)->flags & CFLGS_OBJFREELIST_SLAB)
Linus Torvalds's avatar
Linus Torvalds committed
257
258
259
#define	OFF_SLAB(x)	((x)->flags & CFLGS_OFF_SLAB)

#define BATCHREFILL_LIMIT	16
Andrew Morton's avatar
Andrew Morton committed
260
261
262
/*
 * Optimization question: fewer reaps means less probability for unnessary
 * cpucache drain/refill cycles.
Linus Torvalds's avatar
Linus Torvalds committed
263
 *
Adrian Bunk's avatar
Adrian Bunk committed
264
 * OTOH the cpuarrays can contain lots of objects,
Linus Torvalds's avatar
Linus Torvalds committed
265
266
 * which could lock up otherwise freeable slabs.
 */
267
268
#define REAPTIMEOUT_AC		(2*HZ)
#define REAPTIMEOUT_NODE	(4*HZ)
Linus Torvalds's avatar
Linus Torvalds committed
269
270
271
272
273
274

#if STATS
#define	STATS_INC_ACTIVE(x)	((x)->num_active++)
#define	STATS_DEC_ACTIVE(x)	((x)->num_active--)
#define	STATS_INC_ALLOCED(x)	((x)->num_allocations++)
#define	STATS_INC_GROWN(x)	((x)->grown++)
275
#define	STATS_ADD_REAPED(x,y)	((x)->reaped += (y))
Andrew Morton's avatar
Andrew Morton committed
276
277
278
279
280
#define	STATS_SET_HIGH(x)						\
	do {								\
		if ((x)->num_active > (x)->high_mark)			\
			(x)->high_mark = (x)->num_active;		\
	} while (0)
Linus Torvalds's avatar
Linus Torvalds committed
281
282
#define	STATS_INC_ERR(x)	((x)->errors++)
#define	STATS_INC_NODEALLOCS(x)	((x)->node_allocs++)
283
#define	STATS_INC_NODEFREES(x)	((x)->node_frees++)
284
#define STATS_INC_ACOVERFLOW(x)   ((x)->node_overflow++)
Andrew Morton's avatar
Andrew Morton committed
285
286
287
288
289
#define	STATS_SET_FREEABLE(x, i)					\
	do {								\
		if ((x)->max_freeable < i)				\
			(x)->max_freeable = i;				\
	} while (0)
Linus Torvalds's avatar
Linus Torvalds committed
290
291
292
293
294
295
296
297
298
#define STATS_INC_ALLOCHIT(x)	atomic_inc(&(x)->allochit)
#define STATS_INC_ALLOCMISS(x)	atomic_inc(&(x)->allocmiss)
#define STATS_INC_FREEHIT(x)	atomic_inc(&(x)->freehit)
#define STATS_INC_FREEMISS(x)	atomic_inc(&(x)->freemiss)
#else
#define	STATS_INC_ACTIVE(x)	do { } while (0)
#define	STATS_DEC_ACTIVE(x)	do { } while (0)
#define	STATS_INC_ALLOCED(x)	do { } while (0)
#define	STATS_INC_GROWN(x)	do { } while (0)
299
#define	STATS_ADD_REAPED(x,y)	do { (void)(y); } while (0)
Linus Torvalds's avatar
Linus Torvalds committed
300
301
302
#define	STATS_SET_HIGH(x)	do { } while (0)
#define	STATS_INC_ERR(x)	do { } while (0)
#define	STATS_INC_NODEALLOCS(x)	do { } while (0)
303
#define	STATS_INC_NODEFREES(x)	do { } while (0)
304
#define STATS_INC_ACOVERFLOW(x)   do { } while (0)
Andrew Morton's avatar
Andrew Morton committed
305
#define	STATS_SET_FREEABLE(x, i) do { } while (0)
Linus Torvalds's avatar
Linus Torvalds committed
306
307
308
309
310
311
312
313
#define STATS_INC_ALLOCHIT(x)	do { } while (0)
#define STATS_INC_ALLOCMISS(x)	do { } while (0)
#define STATS_INC_FREEHIT(x)	do { } while (0)
#define STATS_INC_FREEMISS(x)	do { } while (0)
#endif

#if DEBUG

Andrew Morton's avatar
Andrew Morton committed
314
315
/*
 * memory layout of objects:
Linus Torvalds's avatar
Linus Torvalds committed
316
 * 0		: objp
317
 * 0 .. cachep->obj_offset - BYTES_PER_WORD - 1: padding. This ensures that
Linus Torvalds's avatar
Linus Torvalds committed
318
319
 * 		the end of an object is aligned with the end of the real
 * 		allocation. Catches writes behind the end of the allocation.
320
 * cachep->obj_offset - BYTES_PER_WORD .. cachep->obj_offset - 1:
Linus Torvalds's avatar
Linus Torvalds committed
321
 * 		redzone word.
322
 * cachep->obj_offset: The real object.
323
324
 * cachep->size - 2* BYTES_PER_WORD: redzone word [BYTES_PER_WORD long]
 * cachep->size - 1* BYTES_PER_WORD: last caller address
Andrew Morton's avatar
Andrew Morton committed
325
 *					[BYTES_PER_WORD long]
Linus Torvalds's avatar
Linus Torvalds committed
326
 */
327
static int obj_offset(struct kmem_cache *cachep)
Linus Torvalds's avatar
Linus Torvalds committed
328
{
329
	return cachep->obj_offset;
Linus Torvalds's avatar
Linus Torvalds committed
330
331
}

332
static unsigned long long *dbg_redzone1(struct kmem_cache *cachep, void *objp)
Linus Torvalds's avatar
Linus Torvalds committed
333
334
{
	BUG_ON(!(cachep->flags & SLAB_RED_ZONE));
335
336
	return (unsigned long long*) (objp + obj_offset(cachep) -
				      sizeof(unsigned long long));
Linus Torvalds's avatar
Linus Torvalds committed
337
338
}

339
static unsigned long long *dbg_redzone2(struct kmem_cache *cachep, void *objp)
Linus Torvalds's avatar
Linus Torvalds committed
340
341
342
{
	BUG_ON(!(cachep->flags & SLAB_RED_ZONE));
	if (cachep->flags & SLAB_STORE_USER)
343
		return (unsigned long long *)(objp + cachep->size -
344
					      sizeof(unsigned long long) -
David Woodhouse's avatar
David Woodhouse committed
345
					      REDZONE_ALIGN);
346
	return (unsigned long long *) (objp + cachep->size -
347
				       sizeof(unsigned long long));
Linus Torvalds's avatar
Linus Torvalds committed
348
349
}

350
static void **dbg_userword(struct kmem_cache *cachep, void *objp)
Linus Torvalds's avatar
Linus Torvalds committed
351
352
{
	BUG_ON(!(cachep->flags & SLAB_STORE_USER));
353
	return (void **)(objp + cachep->size - BYTES_PER_WORD);
Linus Torvalds's avatar
Linus Torvalds committed
354
355
356
357
}

#else

358
#define obj_offset(x)			0
359
360
#define dbg_redzone1(cachep, objp)	({BUG(); (unsigned long long *)NULL;})
#define dbg_redzone2(cachep, objp)	({BUG(); (unsigned long long *)NULL;})
Linus Torvalds's avatar
Linus Torvalds committed
361
362
363
364
#define dbg_userword(cachep, objp)	({BUG(); (void **)NULL;})

#endif

365
366
#ifdef CONFIG_DEBUG_SLAB_LEAK

367
static inline bool is_store_user_clean(struct kmem_cache *cachep)
368
{
369
370
	return atomic_read(&cachep->store_user_clean) == 1;
}
371

372
373
374
375
static inline void set_store_user_clean(struct kmem_cache *cachep)
{
	atomic_set(&cachep->store_user_clean, 1);
}
376

377
378
379
380
static inline void set_store_user_dirty(struct kmem_cache *cachep)
{
	if (is_store_user_clean(cachep))
		atomic_set(&cachep->store_user_clean, 0);
381
382
383
}

#else
384
static inline void set_store_user_dirty(struct kmem_cache *cachep) {}
385
386
387

#endif

Linus Torvalds's avatar
Linus Torvalds committed
388
/*
389
390
 * Do not go above this order unless 0 objects fit into the slab or
 * overridden on the command line.
Linus Torvalds's avatar
Linus Torvalds committed
391
 */
392
393
394
#define	SLAB_MAX_ORDER_HI	1
#define	SLAB_MAX_ORDER_LO	0
static int slab_max_order = SLAB_MAX_ORDER_LO;
395
static bool slab_max_order_set __initdata;
Linus Torvalds's avatar
Linus Torvalds committed
396

397
398
static inline struct kmem_cache *virt_to_cache(const void *obj)
{
399
	struct page *page = virt_to_head_page(obj);
400
	return page->slab_cache;
401
402
}

403
static inline void *index_to_obj(struct kmem_cache *cache, struct page *page,
404
405
				 unsigned int idx)
{
406
	return page->s_mem + cache->size * idx;
407
408
}

409
/*
410
411
412
 * We want to avoid an expensive divide : (offset / cache->size)
 *   Using the fact that size is a constant for a particular cache,
 *   we can replace (offset / cache->size) by
413
414
415
 *   reciprocal_divide(offset, cache->reciprocal_buffer_size)
 */
static inline unsigned int obj_to_index(const struct kmem_cache *cache,
416
					const struct page *page, void *obj)
417
{
418
	u32 offset = (obj - page->s_mem);
419
	return reciprocal_divide(offset, cache->reciprocal_buffer_size);
420
421
}

422
#define BOOT_CPUCACHE_ENTRIES	1
Linus Torvalds's avatar
Linus Torvalds committed
423
/* internal cache of cache description objs */
424
static struct kmem_cache kmem_cache_boot = {
425
426
427
	.batchcount = 1,
	.limit = BOOT_CPUCACHE_ENTRIES,
	.shared = 1,
428
	.size = sizeof(struct kmem_cache),
429
	.name = "kmem_cache",
Linus Torvalds's avatar
Linus Torvalds committed
430
431
};

432
static DEFINE_PER_CPU(struct delayed_work, slab_reap_work);
Linus Torvalds's avatar
Linus Torvalds committed
433

434
static inline struct array_cache *cpu_cache_get(struct kmem_cache *cachep)
Linus Torvalds's avatar
Linus Torvalds committed
435
{
436
	return this_cpu_ptr(cachep->cpu_cache);
Linus Torvalds's avatar
Linus Torvalds committed
437
438
}

Andrew Morton's avatar
Andrew Morton committed
439
440
441
/*
 * Calculate the number of objects and left-over bytes for a given buffer size.
 */
442
static unsigned int cache_estimate(unsigned long gfporder, size_t buffer_size,
443
		slab_flags_t flags, size_t *left_over)
444
{
445
	unsigned int num;
446
	size_t slab_size = PAGE_SIZE << gfporder;
Linus Torvalds's avatar
Linus Torvalds committed
447

448
449
450
451
452
453
	/*
	 * The slab management structure can be either off the slab or
	 * on it. For the latter case, the memory allocated for a
	 * slab is used for:
	 *
	 * - @buffer_size bytes for each object
454
455
456
457
458
	 * - One freelist_idx_t for each object
	 *
	 * We don't need to consider alignment of freelist because
	 * freelist will be at the end of slab page. The objects will be
	 * at the correct alignment.
459
460
461
462
463
464
	 *
	 * If the slab management structure is off the slab, then the
	 * alignment will already be calculated into the size. Because
	 * the slabs are all pages aligned, the objects will be at the
	 * correct alignment when allocated.
	 */
465
	if (flags & (CFLGS_OBJFREELIST_SLAB | CFLGS_OFF_SLAB)) {
466
		num = slab_size / buffer_size;
467
		*left_over = slab_size % buffer_size;
468
	} else {
469
		num = slab_size / (buffer_size + sizeof(freelist_idx_t));
470
471
		*left_over = slab_size %
			(buffer_size + sizeof(freelist_idx_t));
472
	}
473
474

	return num;
Linus Torvalds's avatar
Linus Torvalds committed
475
476
}

477
#if DEBUG
478
#define slab_error(cachep, msg) __slab_error(__func__, cachep, msg)
Linus Torvalds's avatar
Linus Torvalds committed
479

Andrew Morton's avatar
Andrew Morton committed
480
481
static void __slab_error(const char *function, struct kmem_cache *cachep,
			char *msg)
Linus Torvalds's avatar
Linus Torvalds committed
482
{
483
	pr_err("slab error in %s(): cache `%s': %s\n",
484
	       function, cachep->name, msg);
Linus Torvalds's avatar
Linus Torvalds committed
485
	dump_stack();
486
	add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
Linus Torvalds's avatar
Linus Torvalds committed
487
}
488
#endif
Linus Torvalds's avatar
Linus Torvalds committed
489

490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
/*
 * By default on NUMA we use alien caches to stage the freeing of
 * objects allocated from other nodes. This causes massive memory
 * inefficiencies when using fake NUMA setup to split memory into a
 * large number of small nodes, so it can be disabled on the command
 * line
  */

static int use_alien_caches __read_mostly = 1;
static int __init noaliencache_setup(char *s)
{
	use_alien_caches = 0;
	return 1;
}
__setup("noaliencache", noaliencache_setup);

506
507
508
509
510
511
512
513
514
515
516
static int __init slab_max_order_setup(char *str)
{
	get_option(&str, &slab_max_order);
	slab_max_order = slab_max_order < 0 ? 0 :
				min(slab_max_order, MAX_ORDER - 1);
	slab_max_order_set = true;

	return 1;
}
__setup("slab_max_order=", slab_max_order_setup);

517
518
519
520
521
522
523
#ifdef CONFIG_NUMA
/*
 * Special reaping functions for NUMA systems called from cache_reap().
 * These take care of doing round robin flushing of alien caches (containing
 * objects freed on different nodes from which they were allocated) and the
 * flushing of remote pcps by calling drain_node_pages.
 */
524
static DEFINE_PER_CPU(unsigned long, slab_reap_node);
525
526
527

static void init_reap_node(int cpu)
{
528
529
	per_cpu(slab_reap_node, cpu) = next_node_in(cpu_to_mem(cpu),
						    node_online_map);
530
531
532
533
}

static void next_reap_node(void)
{
534
	int node = __this_cpu_read(slab_reap_node);
535

536
	node = next_node_in(node, node_online_map);
537
	__this_cpu_write(slab_reap_node, node);
538
539
540
541
542
543
544
}

#else
#define init_reap_node(cpu) do { } while (0)
#define next_reap_node(void) do { } while (0)
#endif

Linus Torvalds's avatar
Linus Torvalds committed
545
546
547
548
549
550
551
/*
 * Initiate the reap timer running on the target CPU.  We run at around 1 to 2Hz
 * via the workqueue/eventd.
 * Add the CPU number into the expiration time to minimize the possibility of
 * the CPUs getting into lockstep and contending for the global cache chain
 * lock.
 */
552
static void start_cpu_timer(int cpu)
Linus Torvalds's avatar
Linus Torvalds committed
553
{
554
	struct delayed_work *reap_work = &per_cpu(slab_reap_work, cpu);
Linus Torvalds's avatar
Linus Torvalds committed
555

556
	if (reap_work->work.func == NULL) {
557
		init_reap_node(cpu);
558
		INIT_DEFERRABLE_WORK(reap_work, cache_reap);
559
560
		schedule_delayed_work_on(cpu, reap_work,
					__round_jiffies_relative(HZ, cpu));
Linus Torvalds's avatar
Linus Torvalds committed
561
562
563
	}
}

564
static void init_arraycache(struct array_cache *ac, int limit, int batch)
Linus Torvalds's avatar
Linus Torvalds committed
565
{
566
567
	/*
	 * The array_cache structures contain pointers to free object.
Lucas De Marchi's avatar
Lucas De Marchi committed
568
	 * However, when such objects are allocated or transferred to another
569
570
571
572
	 * cache the pointers are not cleared and they could be counted as
	 * valid references during a kmemleak scan. Therefore, kmemleak must
	 * not scan such objects.
	 */
573
574
575
576
577
578
	kmemleak_no_scan(ac);
	if (ac) {
		ac->avail = 0;
		ac->limit = limit;
		ac->batchcount = batch;
		ac->touched = 0;
Linus Torvalds's avatar
Linus Torvalds committed
579
	}
580
581
582
583
584
}

static struct array_cache *alloc_arraycache(int node, int entries,
					    int batchcount, gfp_t gfp)
{
585
	size_t memsize = sizeof(void *) * entries + sizeof(struct array_cache);
586
587
588
589
590
	struct array_cache *ac = NULL;

	ac = kmalloc_node(memsize, gfp, node);
	init_arraycache(ac, entries, batchcount);
	return ac;
Linus Torvalds's avatar
Linus Torvalds committed
591
592
}

593
594
static noinline void cache_free_pfmemalloc(struct kmem_cache *cachep,
					struct page *page, void *objp)
595
{
596
597
598
	struct kmem_cache_node *n;
	int page_node;
	LIST_HEAD(list);
599

600
601
	page_node = page_to_nid(page);
	n = get_node(cachep, page_node);
602

603
604
605
	spin_lock(&n->list_lock);
	free_block(cachep, &objp, 1, page_node, &list);
	spin_unlock(&n->list_lock);
606

607
	slabs_destroy(cachep, &list);
608
609
}

610
611
612
613
614
615
616
617
618
619
/*
 * Transfer objects in one arraycache to another.
 * Locking must be handled by the caller.
 *
 * Return the number of entries transferred.
 */
static int transfer_objects(struct array_cache *to,
		struct array_cache *from, unsigned int max)
{
	/* Figure out how many entries to transfer */
620
	int nr = min3(from->avail, max, to->limit - to->avail);
621
622
623
624
625
626
627
628
629
630
631
632

	if (!nr)
		return 0;

	memcpy(to->entry + to->avail, from->entry + from->avail -nr,
			sizeof(void *) *nr);

	from->avail -= nr;
	to->avail += nr;
	return nr;
}

633
634
635
#ifndef CONFIG_NUMA

#define drain_alien_cache(cachep, alien) do { } while (0)
636
#define reap_alien(cachep, n) do { } while (0)
637

Joonsoo Kim's avatar
Joonsoo Kim committed
638
639
static inline struct alien_cache **alloc_alien_cache(int node,
						int limit, gfp_t gfp)
640
{
641
	return NULL;
642
643
}

Joonsoo Kim's avatar
Joonsoo Kim committed
644
static inline void free_alien_cache(struct alien_cache **ac_ptr)
645
646
647
648
649
650
651
652
653
654
655
656
657
658
{
}

static inline int cache_free_alien(struct kmem_cache *cachep, void *objp)
{
	return 0;
}

static inline void *alternate_node_alloc(struct kmem_cache *cachep,
		gfp_t flags)
{
	return NULL;
}

659
static inline void *____cache_alloc_node(struct kmem_cache *cachep,
660
661
662
663
664
		 gfp_t flags, int nodeid)
{
	return NULL;
}

David Rientjes's avatar
David Rientjes committed
665
666
static inline gfp_t gfp_exact_node(gfp_t flags)
{
667
	return flags & ~__GFP_NOFAIL;
David Rientjes's avatar
David Rientjes committed
668
669
}

670
671
#else	/* CONFIG_NUMA */

672
static void *____cache_alloc_node(struct kmem_cache *, gfp_t, int);
673
static void *alternate_node_alloc(struct kmem_cache *, gfp_t);
674

Joonsoo Kim's avatar
Joonsoo Kim committed
675
676
677
static struct alien_cache *__alloc_alien_cache(int node, int entries,
						int batch, gfp_t gfp)
{
678
	size_t memsize = sizeof(void *) * entries + sizeof(struct alien_cache);
Joonsoo Kim's avatar
Joonsoo Kim committed
679
680
681
682
	struct alien_cache *alc = NULL;

	alc = kmalloc_node(memsize, gfp, node);
	init_arraycache(&alc->ac, entries, batch);
683
	spin_lock_init(&alc->lock);
Joonsoo Kim's avatar
Joonsoo Kim committed
684
685
686
687
	return alc;
}

static struct alien_cache **alloc_alien_cache(int node, int limit, gfp_t gfp)
688
{
Joonsoo Kim's avatar
Joonsoo Kim committed
689
	struct alien_cache **alc_ptr;
690
	size_t memsize = sizeof(void *) * nr_node_ids;
691
692
693
694
	int i;

	if (limit > 1)
		limit = 12;
Joonsoo Kim's avatar
Joonsoo Kim committed
695
696
697
698
699
700
701
702
703
704
705
706
707
	alc_ptr = kzalloc_node(memsize, gfp, node);
	if (!alc_ptr)
		return NULL;

	for_each_node(i) {
		if (i == node || !node_online(i))
			continue;
		alc_ptr[i] = __alloc_alien_cache(node, limit, 0xbaadf00d, gfp);
		if (!alc_ptr[i]) {
			for (i--; i >= 0; i--)
				kfree(alc_ptr[i]);
			kfree(alc_ptr);
			return NULL;
708
709
		}
	}
Joonsoo Kim's avatar
Joonsoo Kim committed
710
	return alc_ptr;
711
712
}

Joonsoo Kim's avatar
Joonsoo Kim committed
713
static void free_alien_cache(struct alien_cache **alc_ptr)
714
715
716
{
	int i;

Joonsoo Kim's avatar
Joonsoo Kim committed
717
	if (!alc_ptr)
718
719
		return;
	for_each_node(i)
Joonsoo Kim's avatar
Joonsoo Kim committed
720
721
	    kfree(alc_ptr[i]);
	kfree(alc_ptr);
722
723
}

724
static void __drain_alien_cache(struct kmem_cache *cachep,
725
726
				struct array_cache *ac, int node,
				struct list_head *list)
727
{
728
	struct kmem_cache_node *n = get_node(cachep, node);
729
730

	if (ac->avail) {
731
		spin_lock(&n->list_lock);
732
733
734
735
736
		/*
		 * Stuff objects into the remote nodes shared array first.
		 * That way we could avoid the overhead of putting the objects
		 * into the free lists and getting them back later.
		 */
737
738
		if (n->shared)
			transfer_objects(n->shared, ac, ac->limit);
739

740
		free_block(cachep, ac->entry, ac->avail, node, list);
741
		ac->avail = 0;
742
		spin_unlock(&n->list_lock);
743
744
745
	}
}

746
747
748
/*
 * Called from cache_reap() to regularly drain alien caches round robin.
 */
749
static void reap_alien(struct kmem_cache *cachep, struct kmem_cache_node *n)
750
{
751
	int node = __this_cpu_read(slab_reap_node);
752

753
	if (n->alien) {
Joonsoo Kim's avatar
Joonsoo Kim committed
754
755
756
757
758
		struct alien_cache *alc = n->alien[node];
		struct array_cache *ac;

		if (alc) {
			ac = &alc->ac;
759
			if (ac->avail && spin_trylock_irq(&alc->lock)) {
760
761
762
				LIST_HEAD(list);

				__drain_alien_cache(cachep, ac, node, &list);
763
				spin_unlock_irq(&alc->lock);
764
				slabs_destroy(cachep, &list);
Joonsoo Kim's avatar
Joonsoo Kim committed
765
			}
766
767
768
769
		}
	}
}

Andrew Morton's avatar
Andrew Morton committed
770
static void drain_alien_cache(struct kmem_cache *cachep,
Joonsoo Kim's avatar
Joonsoo Kim committed
771
				struct alien_cache **alien)
772
{
773
	int i = 0;
Joonsoo Kim's avatar
Joonsoo Kim committed
774
	struct alien_cache *alc;
775
776
777
778
	struct array_cache *ac;
	unsigned long flags;

	for_each_online_node(i) {
Joonsoo Kim's avatar
Joonsoo Kim committed
779
780
		alc = alien[i];
		if (alc) {
781
782
			LIST_HEAD(list);

Joonsoo Kim's avatar
Joonsoo Kim committed
783
			ac = &alc->ac;
784
			spin_lock_irqsave(&alc->lock, flags);
785
			__drain_alien_cache(cachep, ac, i, &list);
786
			spin_unlock_irqrestore(&alc->lock, flags);
787
			slabs_destroy(cachep, &list);
788
789
790
		}
	}
}
791

792
793
static int __cache_free_alien(struct kmem_cache *cachep, void *objp,
				int node, int page_node)
794
{
795
	struct kmem_cache_node *n;
Joonsoo Kim's avatar
Joonsoo Kim committed
796
797
	struct alien_cache *alien = NULL;
	struct array_cache *ac;
798
	LIST_HEAD(list);
799

800
	n = get_node(cachep, node);
801
	STATS_INC_NODEFREES(cachep);
802
803
	if (n->alien && n->alien[page_node]) {
		alien = n->alien[page_node];
Joonsoo Kim's avatar
Joonsoo Kim committed
804
		ac = &alien->ac;
805
		spin_lock(&alien->lock);
Joonsoo Kim's avatar
Joonsoo Kim committed
806
		if (unlikely(ac->avail == ac->limit)) {
807
			STATS_INC_ACOVERFLOW(cachep);
808
			__drain_alien_cache(cachep, ac, page_node, &list);
809
		}
810
		ac->entry[ac->avail++] = objp;
811
		spin_unlock(&alien->lock);
812
		slabs_destroy(cachep, &list);
813
	} else {
814
		n = get_node(cachep, page_node);
815
		spin_lock(&n->list_lock);
816
		free_block(cachep, &objp, 1, page_node, &list);
817
		spin_unlock(&n->list_lock);
818
		slabs_destroy(cachep, &list);
819
820
821
	}
	return 1;
}
822
823
824
825
826
827
828
829
830
831
832
833
834
835

static inline int cache_free_alien(struct kmem_cache *cachep, void *objp)
{
	int page_node = page_to_nid(virt_to_page(objp));
	int node = numa_mem_id();
	/*
	 * Make sure we are not freeing a object from another node to the array
	 * cache on this cpu.
	 */
	if (likely(node == page_node))
		return 0;

	return __cache_free_alien(cachep, objp, node, page_node);
}
David Rientjes's avatar
David Rientjes committed
836
837

/*
838
839
 * Construct gfp mask to allocate from a specific node but do not reclaim or
 * warn about failures.
David Rientjes's avatar
David Rientjes committed
840
841
842
 */
static inline gfp_t gfp_exact_node(gfp_t flags)
{
843
	return (flags | __GFP_THISNODE | __GFP_NOWARN) & ~(__GFP_RECLAIM|__GFP_NOFAIL);
David Rientjes's avatar
David Rientjes committed
844
}
845
846
#endif

847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
static int init_cache_node(struct kmem_cache *cachep, int node, gfp_t gfp)
{
	struct kmem_cache_node *n;

	/*
	 * Set up the kmem_cache_node for cpu before we can
	 * begin anything. Make sure some other cpu on this
	 * node has not already allocated this
	 */
	n = get_node(cachep, node);
	if (n) {
		spin_lock_irq(&n->list_lock);
		n->free_limit = (1 + nr_cpus_node(node)) * cachep->batchcount +
				cachep->num;
		spin_unlock_irq(&n->list_lock);

		return 0;
	}

	n = kmalloc_node(sizeof(struct kmem_cache_node), gfp, node);
	if (!n)
		return -ENOMEM;

	kmem_cache_node_init(n);
	n->next_reap = jiffies + REAPTIMEOUT_NODE +
		    ((unsigned long)cachep) % REAPTIMEOUT_NODE;

	n->free_limit =
		(1 + nr_cpus_node(node)) * cachep->batchcount + cachep->num;

	/*
	 * The kmem_cache_nodes don't come and go as CPUs
	 * come and go.  slab_mutex is sufficient
	 * protection here.
	 */
	cachep->node[node] = n;

	return 0;
}

887
#if (defined(CONFIG_NUMA) && defined(CONFIG_MEMORY_HOTPLUG)) || defined(CONFIG_SMP)
888
/*
889
 * Allocates and initializes node for a node on each slab cache, used for
890
 * either memory or cpu hotplug.  If memory is being hot-added, the kmem_cache_node
891
 * will be allocated off-node since memory is not yet online for the new node.
892
 * When hotplugging memory or a cpu, existing node are not replaced if
893
894
 * already in use.
 *
895
 * Must hold slab_mutex.
896
 */
897
static int init_cache_node_node(int node)
898
{
899
	int ret;
900
901
	struct kmem_cache *cachep;

902
	list_for_each_entry(cachep, &slab_caches, list) {
903
904
905
		ret = init_cache_node(cachep, node, GFP_KERNEL);
		if (ret)
			return ret;
906
	}
907

908
909
	return 0;
}
910
#endif
911

912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
static int setup_kmem_cache_node(struct kmem_cache *cachep,
				int node, gfp_t gfp, bool force_change)
{
	int ret = -ENOMEM;
	struct kmem_cache_node *n;
	struct array_cache *old_shared = NULL;
	struct array_cache *new_shared = NULL;
	struct alien_cache **new_alien = NULL;
	LIST_HEAD(list);

	if (use_alien_caches) {
		new_alien = alloc_alien_cache(node, cachep->limit, gfp);
		if (!new_alien)
			goto fail;
	}

	if (cachep->shared) {
		new_shared = alloc_arraycache(node,
			cachep->shared * cachep->batchcount, 0xbaadf00d, gfp);
		if (!new_shared)
			goto fail;
	}

	ret = init_cache_node(cachep, node, gfp);
	if (ret)
		goto fail;

	n = get_node(cachep, node);
	spin_lock_irq(&n->list_lock);
	if (n->shared && force_change) {
		free_block(cachep, n->shared->entry,
				n->shared->avail, node, &list);
		n->shared->avail = 0;
	}

	if (!n->shared || force_change) {
		old_shared = n->shared;
		n->shared = new_shared;
		new_shared = NULL;
	}

	if (!n->alien) {
		n->alien = new_alien;
		new_alien = NULL;
	}

	spin_unlock_irq(&n->list_lock);
	slabs_destroy(cachep, &list);

961
962
963
964
965
966
	/*
	 * To protect lockless access to n->shared during irq disabled context.
	 * If n->shared isn't NULL in irq disabled context, accessing to it is
	 * guaranteed to be valid until irq is re-enabled, because it will be
	 * freed after synchronize_sched().
	 */
967
	if (old_shared && force_change)
968
969
		synchronize_sched();

970
971
972
973
974
975
976
977
fail:
	kfree(old_shared);
	kfree(new_shared);
	free_alien_cache(new_alien);

	return ret;
}

978
979
#ifdef CONFIG_SMP

980
static void cpuup_canceled(long cpu)
981
982
{
	struct kmem_cache *cachep;
983
	struct kmem_cache_node *n = NULL;
984
	int node = cpu_to_mem(cpu);
985
	const struct cpumask *mask = cpumask_of_node(node);
986

987
	list_for_each_entry(cachep, &slab_caches, list) {
988
989
		struct array_cache *nc;
		struct array_cache *shared;
Joonsoo Kim's avatar
Joonsoo Kim committed
990
		struct alien_cache **alien;
991
		LIST_HEAD(list);
992

993
		n = get_node(cachep, node);
994
		if (!n)
995
			continue;
996

997
		spin_lock_irq(&n->list_lock);
998

999
1000
		/* Free limit for this kmem_cache_node */
		n->free_limit -= cachep->batchcount;
1001
1002
1003
1004

		/* cpu is dead; no one can alloc from it. */
		nc = per_cpu_ptr(cachep->cpu_cache, cpu);
		if (nc) {
1005
			free_block(cachep, nc->entry, nc->avail, node, &list);
1006
1007
			nc->avail = 0;
		}
1008

1009
		if (!cpumask_empty(mask)) {
1010
			spin_unlock_irq(&n->list_lock);
1011
			goto free_slab;
1012
1013
		}

1014
		shared = n->shared;
1015
1016
		if (shared) {
			free_block(cachep, shared->entry,
1017
				   shared->avail, node, &list);
1018
			n->shared = NULL;
1019
1020
		}

1021
1022
		alien = n->alien;
		n->alien = NULL;
1023

1024
		spin_unlock_irq(&n->list_lock);
1025
1026
1027
1028
1029
1030

		kfree(shared);
		if (alien) {
			drain_alien_cache(cachep, alien);
			free_alien_cache(alien);
		}
1031
1032

free_slab:
1033
		slabs_destroy(cachep, &list);
1034
1035
1036
1037
1038
1039
	}
	/*
	 * In the previous loop, all the objects were freed to
	 * the respective cache's slabs,  now we can go ahead and
	 * shrink each nodelist to its limit.
	 */
1040
	list_for_each_entry(cachep, &slab_caches, list) {
1041
		n = get_node(cachep, node);
1042
		if (!n)
1043
			continue;
1044
		drain_freelist(cachep, n, INT_MAX);
1045
1046
1047
	}
}

1048
static int cpuup_prepare(long cpu)
Linus Torvalds's avatar
Linus Torvalds committed
1049
{
1050
	struct kmem_cache *cachep;
1051
	int node = cpu_to_mem(cpu);
1052
	int err;
Linus Torvalds's avatar
Linus Torvalds committed
1053

1054
1055
1056
1057
	/*
	 * We need to do this right in the beginning since
	 * alloc_arraycache's are going to use this list.
	 * kmalloc_node allows us to add the slab to the right
1058
	 * kmem_cache_node and not this cpu's kmem_cache_node
1059
	 */
1060
	err = init_cache_node_node(node);
1061